Adding to what another user said here, when using LLMs for heavy prototyping, I also commit frequently and have a document outlining the opinions contained in the repo.
One thing I've also been doing now is I have a "template" to use for fullstack applications that's configured out-of-the-box with tech that I like, has linters and hooks set up, as well as is structured in a way that's natural to me. This makes it a lot more likely that the generated code will follow my preferred style. It's also not uncommon that on the first iteration I get some code that's not what I'd like to have in my codebase and I just say "refactor X in this and this way, see <file> for an example" at which point I mostly get code that I'm happy with.
One thing I've also been doing now is I have a "template" to use for fullstack applications that's configured out-of-the-box with tech that I like, has linters and hooks set up, as well as is structured in a way that's natural to me. This makes it a lot more likely that the generated code will follow my preferred style. It's also not uncommon that on the first iteration I get some code that's not what I'd like to have in my codebase and I just say "refactor X in this and this way, see <file> for an example" at which point I mostly get code that I'm happy with.